A useful directory can come before a complete benchmark.
We start with twelve assistant profiles linked to official sources. They help you discover products and understand their positioning, but they do not prove how well a product performs.
A small test, done properly.
Our first proposed round focuses on knowledge work: document synthesis, constrained writing and source verification. We will compare similar products using the same inputs and publish the records, limitations and conditions.
Blank scores are honest scores.
There are currently no published runs and no community ratings. Rather than fill those spaces with simulated numbers, we show the status and plan. Different task coverage also means a universal average would be misleading.
The next round starts with a question.
Use the directory to form a shortlist. Try a task, record what worked and what failed, and submit requests for the products you want tested. Requests are collected for editorial planning, not treated as public ratings.
Read our methodology